The groups were matched for age, ethnicity, and socioeconomic status, but control subjects were of higher intelligence than CD participants (p < .005; ). Compared with control subjects, significantly more CD participants reported regular use of tobacco [χ2(1) = 27.5, p < .001], alcohol [χ2(1) = 8.5, p < .005], and cannabis [χ2(1) = 15.5, p < .001]. Conduct disorder participants also had higher levels of psychopathic (p < .001) and callous-unemotional (p < .001) traits. There were no group differences in self-reported anxiety.
Facial Identity Recognition
There was no group difference in face recognition [t(53) = 1.13, p = .26]. Mean (± SD) score for control subjects = 46.4/54 (± 2.9) and for CD participants = 45.6/54 (± 2.5).
Facial Expression Recognition
The emotion hexagon task results are shown in . The data were not normally distributed and could not be transformed to a normal distribution. Mann-Whitney tests were therefore used to compare groups on the six expressions separately, using an alpha level of .008 (i.e., .05/6) to control for multiple comparisons.
Accuracy of facial expression recognition by group. Relative to control subjects, participants with conduct disorder showed impairments in anger and disgust recognition. **p < .01, ***p < .005.
Relative to control subjects, CD participants showed significant deficits in recognition of anger (Z = −2.70, p = .007, r = .35) and disgust (Z = −2.86, p = .004, r = .37) but not fear, happiness, sadness, or surprise (all p > .1).
The anger and disgust recognition deficits were still significant (both p = .03; r = .30 and r = .31, respectively) after equating groups on estimated IQ (p = .13) by removing eight high IQ control subjects (>118 on the Wechsler Abbreviated Scale of Intelligence). This procedure reduced sample size; thus, an alpha of .05 was used. To rule out the possibility that the group findings were driven by comorbid MDD or ADHD, we omitted participants with these disorders and repeated the analyses. The findings for anger and disgust were significant (both p = .01, r = .33) when participants with MDD were excluded. Following exclusion of participants with ADHD, the disgust deficit was significant (p < .01, r = .35) and the anger deficit was of borderline significance (p = .055, r = .27).
Effect of Psychopathic Traits on Facial Expression Recognition
Given our a priori hypothesis of impairments in fear and sadness recognition in CD participants high in psychopathic traits, we examined the relationship between YPI score and expression recognition by splitting the CD sample using the YPI cutoff of 2.5. shows facial expression recognition accuracy in high and lower psychopathy CD subgroups.
Figure 2 Effect of psychopathic traits on facial expression recognition, considering participants with conduct disorder only. Participants high in psychopathic traits, as measured using total scores on the Youth Psychopathic traits Inventory, showed a specific (more ...)
Comparing these subgroups revealed deficits in recognition of sadness (Z = −2.96, p = .003, r = .53), but no other emotions, in those high in psychopathic traits. This effect was not accounted for by subgroup differences in IQ (p = .48) or Benton Facial Recognition Test performance (p = .60). Repeating these analyses using callous-unemotional traits yielded similar results for sadness (Z = −2.17, p = .03, r = .41).
Only one control scored above the YPI cutoff, preventing us from using the above method to investigate effects of psychopathic traits on facial expression recognition in control subjects. However, neither overall psychopathic nor callous-unemotional traits were significantly correlated with sadness or fear recognition in control subjects using Spearman's rho.
Data were unavailable for eight participants due to technical or experimenter error or inability to tolerate the procedure. Recall score did not differ between groups (p = .95); both could articulate the experimental contingencies.
The SCR data were not normally distributed, so they were normalized using a log(SCR + 1) transformation. Raw values are shown in the figures for ease of interpretation.
We first examined whether SCRs to the US differed over time and by group, using repeated-measures analysis of variance (ANOVA) with time as within-subjects factor and group as between-subjects factor (10 × 2). This revealed effects of time [F(4.61,212.03) = 26.38, p < .001, ηp2 = .37] and group [F(1,46) = 5.18, p < .05, ηp2 = .10] but no interaction (F < 1). Skin conductance responses to the US were lower in CD participants than control subjects (). Additionally, SCRs to the US declined strongly over time in both groups.
Figure 3 Skin conductance responses (SCRs) to the 10 presentations of the aversive unconditioned stimulus, by group. Although both groups showed marked habituation of SCRs to the unconditioned stimulus over time, control subjects showed significantly larger SCRs (more ...)
Group differences in autonomic conditioning were assessed using a mixed-model ANOVA with group (control, CD) as a between-subjects factor and CS type (CS+, CS−) and conditioning phase (HAB, ACQ1, ACQ2, EXT) as within-subjects factors (2 × 2 × 4). This revealed effects of CS type [F(1,46) = 21.01, p < .001, ηp2 = .31] and conditioning phase [F(3,138) = 5.78, p < .005, ηp2 = .11] but not group [F(1,46) = 3.11, p = .08]. Underlying the phase effect, SCRs were larger during HAB than ACQ1 or ACQ2 (both p ≤ .01). Underlying the CS type effect, subjects showed larger SCRs to the CS+ than the CS– (p < .001).
We also observed significant group × CS type [F(1,46) = 3.98, p = .05, ηp2 = .08] and conditioning phase × CS type [F(1.47,67.76) = 4.10, p < .05, ηp2 = .08] interactions. Breaking down the former, control subjects exhibited greater SCR differentiation between CS types than CD participants (), as shown using separate repeated-measures ANOVAs for each group, which revealed an effect of CS type in control subjects [F(1,24) = 31.45, p < .001, ηp2 = .57] but not CD participants [F(1,22) = 2.46, p = .13].
Figure 4 Mean (± SE) skin conductance responses to blue test slides (conditioned stimulus positive unpaired with unconditioned stimulus, solid line and closed symbols) and red control slides (conditioned stimulus negative, dashed line and open symbols) (more ...)
The CS type × phase interaction was driven by a divergent pattern of SCRs to the respective CS types between habituation and acquisition 1 and 2. That is, the difference in SCRs between the CS+ and CS– was much greater during acquisition 1 and 2, relative to habituation. There was, however, no significant group × CS type × phase interaction.
To ensure that group differences in autonomic conditioning were not explained by IQ differences, we ran an analysis of covariance (ANCOVA) with IQ as a covariate. The IQ was a significant covariate, but after accounting for IQ, all results except the effect of CS type remained significant and a main effect of group emerged [F(1,45) = 12.76, p < .001, ηp2 = .22]. To examine the impact of comorbidity, we excluded participants with ADHD and reran the analyses. The group × CS type interaction was not significant but had a similar effect size [F(1,41) = 3.42, p = .07, ηp2 = .08]. Crucially, there was still no effect of CS type in this noncomorbid CD subgroup [F(1,17) = 1.52, p = .23]. The group × CS type interaction was significant [F(1,44) = 4.24, p < .05, ηp2 = .09] following exclusion of participants with MDD.
To investigate the possibility that reduced US responsiveness caused the conditioning impairments, we excluded seven CD participants with small SCRs to the US (mean values < .2 μS across 10 US trials). There was no effect of CS type in the CD subgroup with normal SCRs to the US [F(1,15) = 2.0, p = .17].
In separate ANCOVAs controlling for tobacco, alcohol, or cannabis use, the group × CS type interaction remained significant in each case (all p
< .05, ηp2
≥ .09). Finally, we explored whether psychopathic traits influenced fear conditioning. In an ANCOVA with psychopathic traits as a covariate, this variable was not a significant covariate when considering the entire sample. We also assessed for differences in fear conditioning between the low and high psychopathy CD subgroups (Figure S2 in Supplement 1
). There was no effect of subgroup status (p
= .73) nor a subgroup × CS type interaction (p
= .59). Neither subgroup showed an effect of CS type (both p
Startle Reflex Modulation by Affective Valence
Data were unavailable for seven participants due to technical or experimenter error or parents withholding consent for their children to participate in this experiment.
The startle data were not normally distributed, so they were normalized using a square root transformation. Absolute values are provided in for ease of interpretation.
Figure 5 Mean (± SE) startle reflex magnitudes to a 97-dB acoustic probe when viewing pictures of different affective valence, by group. Compared with control subjects, conduct disorder participants exhibited reduced startle magnitudes (p < .05) (more ...)
We ran a repeated-measures ANOVA with slide category as a within-subjects factor and group as a between-subjects factor (5 × 2). This revealed effects of slide category [F(4,180) = 14.79, p < .001, ηp2 = .25] and group [F(1,45) = 4.61, p < .05, ηp2 = .09] but no interaction effect (p = .09). Pairwise comparisons between slide categories showed reduced startle magnitudes to the acoustic probe when participants viewed positive slides, relative to all negative slides (all p < .005). Startle magnitudes were larger when viewing disgusting and fearful slides compared with neutral slides (both p < .005) and when viewing disgusting compared with sad slides (p < .005). Thus, startle magnitudes were significantly modulated by slide category.
Underlying the group effect, CD participants exhibited lower startle magnitudes across all slide categories (). This effect remained significant (p < .05, ηp2 = .15) following exclusion of participants with average startle responses <2.5 mV (nonresponders to the startle probe ) and following statistical adjustment for differences in IQ and psychopathic traits (both p < .05). Furthermore, it was marginally significant following exclusion of participants with MDD (p = .06, ηp2 = .08) or ADHD (p = .06, ηp2 = .08).
The influence of psychopathy on affective modulation was explored further by splitting the CD sample into high and low psychopathy subgroups. There was no subgroup effect on overall startle magnitudes (p
= .88) or a psychopathy subgroup × slide category interaction (p
= .42; Figure S3 in Supplement 1
). Hence, there was no evidence for reduced affective modulation in the high relative to the low psychopathy CD subgroup.
Finally, group effects on startle magnitude remained significant when controlling for differences in tobacco (p < .05, ηp2 = .11), cannabis (p = .05, ηp2 = .08), or alcohol use (p < .005, ηp2 = .18).